Load control in low voltage level of the electricity
grid using
µ
CHP appliances
M.G.C. Bosman, V. Bakker, A. Molderink, J.L. Hurink, G.J.M. Smit
Department of Electrical Engineering, Mathematics and Computer Science
University of Twente
P.O. Box 217, 7500 AE, Enschede, The Netherlands
m.g.c.bosman@utwente.nl
Abstract—The introduction of µCHP (Combined Heat and Power) appliances and other means of distributed generation causes a shift in the way electricity is produced and consumed. Households themselves produce electricity and deliver the surplus to the grid. In this way, the distributed generation also has implications on the transformers and, thus, on the grid. In this work we study the influence of introducing µCHP appliances on the total load of a group of houses (behind the last transformer). If this load can be controlled, the transformer may be relieved from peak loads. Moreover, a well controlled fleet production can be offered as a Virtual Power Plant to the electricity grid.
In this work we focus on different algorithms to control the fleet and produce a constant electricity output. We assume that produced electricity is consumed as locally as possible (preferably within the household). Produced heat can only be consumed locally. Additionally, heat can be stored in heat stores. Fleet control is achieved by using heat led control algorithms and by specifying as objective how much of the µCHP appliances have to run.
First results show that preferred patterns can be produced by using fleet control. However, as the problem is heat driven, still reasonably large deviations from the objective occur. Several combinations of heat store and fleet control algorithm parameters are considered to match the heat demand and supply.
This work is a first attempt in controlling a fleet and gives a starting point for further research in this area. A certain degree of control can already be established, but for better stability more intelligent algorithms are needed.
Keywords: microgeneration, modelling, algorithm design I. INTRODUCTION
The Dutch electricity grid operator has to control the electricity distribution over the country. This grid origins from the situation, in which electricity is centrally generated in large power plants and in which distribution means distribution from these power plants towards customers. Therefore the grid consists of different voltage levels in order to reduce transportation losses over long distances. For long distances a high voltage is required, such that a lower current is sufficient to result in the demanded power. Transformers are used to switch from the one voltage level to the other. At the demand end of the grid, the voltage level is lower. Here, a set of houses is placed behind a last transformer.
The introduction of µCHP appliances and other means of distributed generation causes a shift in the way electricity is produced and consumed. Consumers themselves can produce electricity more efficiently [3] [2] and, moreover, they can
deliver this electricity to the grid. Of course, this will have implications on the transformers and thus on the grid. The grid is obliged to accept electricity that is delivered to it below a certain amount per year. Since the electricity production of the µCHP is below the kiloWatt level, this amount is probably not reached and all produced electricity has to be accepted by the grid. If the use of µCHP is applied on large scale, problems will arise, since the combined households can produce a lot of electricity at the same time, which causes peak loads in the electricity grid. This means that the total production should be controlled at all time slices.
In this work we study the influence of introducing µCHP appliances on the total load of a group of houses (behind a last transformer). If this load can be controlled, the transformer might be relieved from peak loads. Moreover, a well controlled fleet production can be offered to the electricity grid as the product of a Virtual Power Plant. In the first control method we emphasize on heat led control, which gives priority to support a certain minimum heat demand in each household at each moment in time.
We focus on the last part of the electricity chain and the goal is to control the load on the last transformer (i.e. the production of the group of houses). In Section II-A a description of the model of the grid is given. The objective is explained in more detail in Section II-B. Based on this objective the control method is discussed in Section II-C and simulation scenarios are shown in Section II-D. Results are given in Section III.
II. APPROACH
The goal of this work is to develop µCHP controllers for in house use and fleet controllers that take care of a group of houses. These controllers together have to be able to give a stable grid, which means a controlled maximum load on the transformer and a stable electricity production. To test the developed control methods, a simulator for a group of houses is built [4]. First, we present the most important aspects of the simulator. Then the used control methods and different simulation scenarios are presented.
A. Model
Figure 1 models a group of houses behind a transformer, connected to the electricity grid. The contents of a house are shown in the same figure. A house has several (electrical and
grid transf ormer
house1 house3 house5
house2 house4 house6
µC H P Appliance1 Appliance2 . . . Appliancen2 H eatstor e Controller
Fig. 1. Overview of the model
heat demanding) appliances, one µCHP, one heat store and a local controller. Global control is achieved in the grid (at the transformer). In the following paragraphs the model is explained in more detail.
1) Timing: The model uses a set of discrete time slices
T = {0, . . . , NT}, which results in discrete control
mecha-nisms that decide whether a µCHP appliance is switched on or off for the length of a complete time slice, opposite to other control mechanisms which might work in continuous time.
2) The grid: The model consists of a set of houses H,
which are connected to the grid behind a single transformer. The houses are expected to exchange electricity locally in case of a surplus at one house and a shortage at another house. This implies that the load at the transformer is determined by the difference between the total production and the total demand (during a single time slice) within the grid. Note that the load can be both positive and negative; the sign switches with the direction of the electricity flow. The electricity production of
house i on time j ∈ T is denoted by P Eij and the electricity
demand by DEij. In this work we assume that the electricity
demand data are given and deterministic. The load Lj on time
slice j is defined as being the difference between generation and demand:
Lj=
X
i∈H
(P Eij− DEij)
This value is equal to the imported (in case of a negative load) or exported (in case of a positive load) electricity through the transformer at time j. The maximum load on the transformer
is the maximum absolute Lj value during the planning period:
Lmax= max
j∈T |Lj|
A fleet controller can be used to control the electricity pro-duction within the grid via influencing the number of µCHP appliances that are switched on in each time slice.
A factor, which can influence the control, is the electricity market. This market is modelled as part of the grid and, in the current implementation, uses average day ahead data to give an indication of the electricity prices during a day.
3) The house: A house i ∈ H is modelled by a set of
appliances Ai= {ai0, . . . , aini}, for which heat and electricity
profiles are given. We denote by DEijk (DHijk) the given
electricity (heat) demand of appliance aik in time slice j. The
electricity (heat) demand of a house DEij (DHij) is given by
taking the sum of demands DEijk(DHijk) for all appliances
in Ai.
Furthermore, a µCHP is installed in each house, together with a heat store. If the µCHP is running it produces a certain heat output and a certain electricity output. The heat output is stored in the heat store (resulting in an increased heat level in the heat store) or directly consumed by appliances with heat
demand. Thus, next to an electricity production P Eij we also
get a heat production P Hij for house i and time slice j. The
electricity output is either used by the electricity demand of appliances or exported to the grid.
The µCHP has a controller that decides whether the µCHP is switched on or off for a complete time slice. The binary
decision variable xij represents the on and off status of the
µCHP of house i at time slice j. This decision is based on the energy level of the heat store and possibly also on the heat and electricity profiles and the prices on the electricity market. In
the current implementation only the heat level hi in the heat
store of house i (with capacity hci) is taken into account. An
upper level U L and a lower level LL on the energy level of the heat store are defined to determine the heat levels at which the µCHP must be switched off and on. If the µCHP is switched on it takes some time to work on full power. During this start up time the electricity production is lower than the production at full power, which is 1000 W in this work. The assumption is made that the electricity production increases linearly until it has reached its full power. The same holds for switching off the µCHP appliance (i.e. a linear decrease). Furthermore, due to technical constraints, there is a lower limit on the time the µCHP has to run before it can be switched off again. This minimum run time can have influence on the decision to switch a µCHP on or not. Besides U L and LL another heat level M L is introduced. This value gives the level below which an appliance may be switched on and, therefore, the difference between LL and M L gives the controller some freedom to schedule the electricity production of the fleet. An extra constraint is that when the µCHP is switched off, it has to stay off for a minimum period before it can be turned on again.
A given schedule (defined by xij values) now results in an
electricity (heat) production profile of the fleet, given by the produced kWh per time period.
0 2 4 6 8 10 12 14 16 18 20 22 24 0 1 2 3 4 time (h) heat demand (kW) heat demand
Fleet Controller Hub Hub µCHPµCHPµCHP Hub µCHPµCHPµCHP Hub Hub µCHPµCHPµCHP Hub µCHPµCHPµCHP
Fig. 3. The hierarchical structure of the control mechanism
4) Profiles: In the current implementation each house has
the same appliances and thus the same electricity profile, since all appliances have the same non stochastic profile. The standard heat profile is derived from measurements taken at six houses for the durance of one week in winter [1]. This profile is shown in Figure 2. The heat demand adds up to a total of 52.625 kWh a day, which implies that the µCHP should run for approximately 6.6 hours at full generation mode to supply this demand (without taking startup times and shutting down times into account). The heat profile thus limits to a large extent the scheduling freedom of switching the µCHP on and off. If the minimum run time is half an hour, the startup period is 12 minutes and the switchoff period is 6 minutes, this implies that the µCHP is switched on at most 10 times per day, although it is still more preferable to have as little runs as possible due to wearing of the machine.
5) The controller: The µCHP appliances need to be
con-trolled for several reasons, including the need for control mechanisms for a proper working of the installation and the connection to the in-house architecture. Our focus is on heat driven control, since it is important to guarantee to have a certain amount of heat available whenever there is demand for it in a household. This leads to local µCHP controllers that are suitable to supply the heat demand. However, eventually the most important aspect is to control the electricity grid. For this, we introduce a global controller that guides the local heat led controllers, while working on a global electricity led objective. The global Fleet Controller coordinates the control of the individual µCHP appliances, possibly via different Hubs. These Hubs are used to divide the fleet into groups with a limited number of houses. The lowest Hub is situated at a transformer. The resulting structure of the controllers is shown in Figure 3.
Each controller in the diagram must take the constraints (a certain production pattern) given from the level above into account. Each Hub can decide for itself how it reaches this objective or can split this objective into different parts, which are forwarded to Hubs in lower levels. For this research we consider only one global Fleet Controller and no Hubs.
B. Objective
Before describing the objective, two important aspects of the network are taken into consideration. First, the load of each transformer is of crucial interest. The transformer is adjusted to handle a certain maximum possible load. If this load can be reduced, this has a positive impact on the working of the trans-former. Secondly, the electricity grid needs to be balanced.
It is not possible to simply produce and consume electricity, whenever and wherever wanted. A proper ‘harmony’ between supply and demand is regulated by the electricity market and the grid is controlled by the network operator. Electricity is sold in predetermined batch sizes. If a Fleet Controller wants to act on this market, it has to be able to control the total production of a fleet of houses. Thus, in order to be able to make really use of all the electricity that is produced, the fleet as a whole needs to fullfill demand patterns that are given (e.g. by the Fleet Controller or by a certain Hub). More precisely,
for each time slice j a preferable production amount OEj
(OHj) for the whole fleet is given. Based on these given
amounts OEj or OHj (which are deduced from heat and
electricity predictions, both in households and in the market) we now consider as objective of the fleet controller to:
• minimize the mismatch between the fleet production and
the specified amounts; i.e.
minX j∈T |OEj− X i∈H P Eij| or minX j∈T |OHj− X i∈H P Hij|
This objective is motivated by the structure of the electricity market. In this market a producer (in our case the fleet) has to specify in advance the amounts of electricity it wants to produce. These amounts have to be specified for small time periods (nowadays hours). In a first step we assume that
the amount OEj (and thus OHj, since heat and electricity
production is coupled directly) is constant over the time slices. This means that our choice for the preferable production is independent of the electricity demand of the houses in the fleet. The total electricity demand of the fleet is reduced by a constant; the control method is not aiming at peak reduction. However, for a first scenario this is a proper choice, since it allows to look at possibilities to schedule heat production of individual houses over a day. In later work the preferable production depends on both heat and electricity predictions in order to reduce electricity peaks better. For now, we concentrate on the heat demand.
C. Control method
The objective is to control the production of the fleet in such a way, that heat is generated according to a specified profile
(OHj values), which means in our case at a constant rate.
By using µCHP appliances, there is a fixed ratio between the production of heat and the production of electricity. Since the individual controllers are heat driven, the heat profiles of the houses give a guideline for filling the heat stores. Therefore it is necessary that a control method that is developed for the production of a constant electricity rate, takes the heat store level into account. In general, in case of a higher demand in the heat profile, the µCHP runs longer, since filling the heat store goes slower. Also the time between two runs decreases, since the heat store empties faster.
These characteristics are depicted in Figure 4. If all house-holds are supposed to have more or less the same heat profile (in our case the heat profiles are exactly the same), we observe
4 3 2 1 0 5 10 15 20time (h) heat profile (kWh) time (h) xij 1 0 1 2 3 4 5 time (h) xij 1 10 11 12 13 14 15
Fig. 4. The influence of the heat profile on the µCHP control
problems if we want the fleet to produce the same amount of electricity (and thus heat) in each time period. One way to solve this is to start a busy period (a period in which the heat demand is high) with the heat store as full as possible and to decrease the intermediate time between two runs as much as possible in calm periods. A good choice of LL, M L and U L already gives the opportunity to implement this solution in a simple control method (in which heat comfort is important).
Besides the choices for these threshold values good OHj
values are necessary. These values have to be chosen such that the resulting heat production per day is approximately equal to the total heat demand per day, since we do not want to have large fluctuations in the starting levels of the heat stores for the next day.
The control method works as follows. The preferable
production amounts OHj can be transformed in preferred
numbers ONj of µCHP appliances which should run at full
power in time slice j. The goal is to match this number as close as possible. Therefore the appliance with the lowest level in its corresponding heat store is switched on in case of a shortage of running machines, unless the heat store level is above M L. This level M L is set such that the minimum run time can be achieved in most of the cases without producing heat that cannot be stored for matching demands of appliances. As a consequence of this there is little heat loss. Furthermore, if the level of a µCHP runs below LL, this appliance is switched on, no matter how large the number of appliances currently running is.
In the current method we do not use any forecasting. The heat store level is recorded at the end of a time slice and in the next time slice the controller reacts to this information. However, if the fleet controller knows that in the following time slice a machine is stopping, automatically a new unit may start to produce, if the heat store level allows this.
If for a specific time slice we define R as the number of
running µCHP appliances, hi as the heat store level of house
i and ONj as the preferred number of running machines in
time slice j, the working of the control method is presented in Figure 5.
With parameters NT = 240, |H| = 25, ONj = 7 for
all time slices, LL = 2 kWh, M L = 5.5 kWh, U L = 9
kWh, hi = 5 kWh and hci = 10 kWh set, Figure 6
shows as an example the results of the control method against
Control method stable fleet production
Step 1 Initialization
Set initial values j = 0, R = 0
Step 2 Use Individual House Controllers
For all i ∈ H DO:
If (hi< LL)
Switch µCHP of i on (R := R + 1)
If (hi> U L)
Switch µCHP of i off (R := R − 1)
Step 3 Use Fleet Controller
Receive heat store levels and µCHP states from households
While (R < ONj)
{
If (min
i∈H{hi|µCHP of i off & minimum
off period of i passed} < M L) {
Switch on the µCHP appliance in household i R := R + 1 } Else Go to Step 4 }
Step 4 Update Individual Houses
For all i ∈ H DO:
Update hi
Step 5 Go To Next Time Slice
j := j + 1 If (j < Nt)
Return to Step 2 Else
End
Fig. 5. The control method
0 20 40 60 80 100 120 140 160 180 200 220 240 0 5 10 15 20 time (x 6 minutes) total heat generation (kWh)
individual control fleet control objective heat demand
Fig. 6. Comparison of independent and fleet generation for a fleet of 25 houses (objective: 7 production units running at the same time)
an uncontrolled fleet. Individual control via independently operating households leads to an unbalanced heat (and thus electricity) production, while the control method follows the objective quite well for large parts of the time. However, some deviations, where the fleet cannot fullfill demand or leads to too much production, from this objective still occur. The deviations have two main reasons: shortages are due to heat demand that lags behind at the average preferred production
for a considerable amount of time (all heat stores are ‘full’) and surpluses due to large positive differences between heat demand and preferred heat production (more heat stores need to be filled than preferred). Other small deviations follow from the fact that appliances do not work at full power during the startup and shutting down times, or result from the requirement that some appliances cannot be switched off before the minimum runtime has passed. In the figure also
the total heat demand and the objective OHj values are
shown. The consequences of a positive and negative difference between these values for longer time periods are visible in the production by the fleet control method.
D. Simulation scenario
To show first results of our control method, we propose the following simulation scenario. In this scenario we compare the heat mismatch of using heat led control methods with different parameters, to the mismatch while using independent local controllers.
1) The scenario: We use a fleet of 25 houses. Each house
has a Whispergen µCHP and a Gledhill heat store. The Whis-pergen has a generation level of 1 kW, a minimum runtime of half an hour, a startup time of 12 minutes, a stopping time of 6 minutes and a minimum off period of half an hour. The electricity:heat production ratio of the Whispergen is 1:8. The used heat store has a capacity of 10 kWh and a starting level of 5 kWh. The heat and electricity demanding appliances are the same in all houses and equal user profiles are defined for a single day.
For this fleet the parameters M L, LL and U L are varied. All simulations are run for the durance of a single day. For these scenarios the best match is found by varying the constant
preferable µCHP appliances running ONj from 1 to 25.
In order to find the best combination of heat store
param-eters and objective ONj, the following penalty measure is
defined besides the earlier mentioned heat mismatch:
squared mismatch =X
j∈T
((X
i∈H
(P Hij) − OHj)2)
The ratio squared mismatchmismatch gives a measure for the
oc-curence of mismatches. If this ratio is high, there is probably a small number of time slices in which there is a large difference between production and objective; if the ratio is low, the differences are more spread over the day.
III. RESULTS
In this section we discuss the results of the scenario and give our view at improvements that have to be made in future work.
A. Simulation scenario
Table I shows the results of the scenario, where an uncon-trolled fleet is compared to fleet control. For the fleet control methods we defined the values LL = 1 kWh, M L = 5.5 kWh and U L = 9 kWh as basic parameters. These values are chosen such that every heat demand can still be fullfilled once the µCHP has to be started (LL = 1), that heat loss is avoided
once the µCHP has to start stopping (U L = 9) and that the minimum runtime of half an hour can be achieved before the upper level is reached (M L = 5.5). The influence of changing these parameters with 1 kWh is shown in the table.
Control LL ML UL a) b) c) d) individual 1 5.5 9 7 1833 17529 9.6 fleet 1 5.5 9 7 194 550 2.8 fleet 2 5.5 9 8 253 1026 4.1 fleet 1 5.5 8 7 203 533 2.6 fleet 1 4.5 9 7 292 1105 3.8
a) best objective ONj, b) mismatch
c) squared mismatch, d)squared mismatchmismatch
TABLE I
MISMATCH FOR SIMULATION SCENARIO
The best objective value for ONj is 7. In the basic case
the average mismatch per time slice is 0.8 kWh, which is around 14% of the preferable production per time slice. The uncontrolled fleet has an average mismatch of 136% of the preferred value. The production mismatch can thus be reduced by almost 90% by using fleet control, compared to individual control by households. The squared mismatch is even more reduced, which shows that the fleet control methods result in a more stable production over the length of the day.
The parameter setting of LL and M L seems to be more important than a good value for U L. Lowering the level of U L even gives better results for the squared mismatch. An explanation for this is that the controller gets more freedom, since appliances are switched off earlier. On the contrary, a decrease of the level M L and an increase of LL decreases the decision freedom.
B. Future work
In future work the objective of minimizing the electrical cur-rent at the transformer can be addressed. New fleet controllers should be defined in order to handle electricity peaks. Before this can be looked at, a good prediction of the electricity load is necessary. This is an important aspect of further research. Once electricity usage and prices on the electricity market can be predicted, we can be able to define new non constant
preferable production patterns OHjthat should be reached by
the control method.
IV. ACKNOWLEDGEMENTS
This research is conducted within the SFEER project (07937) supported by STW.
REFERENCES
[1] V. Bakker, A. Molderink, J.L. Hurink, G.J.M. Smit, Domestic heat demand prediction using neural networks, Proceedings of Nineteenth International Conference on Systems Engineering, pp. 189-194, Las Vegas, August 2008
[2] M.P.F. Hommelberg, B. Roossien, C.J. Warmer, J.K. Kok, F.J. Kuijper, J.W. Turkstra, Aggregatie van micro-wkk’s in een virtuele centrale, Public end report first trial smart power system, July 2007
[3] A. de Jong, E.J. Bakker, J. Dam, H. van Wolferen, Technisch energie-en CO2-besparingspotentieel van micro-wkk in Nederland (2010-2030),
Technical report, Werkgroep Decentraal, July 2006
[4] A. Molderink, V. Bakker, M.G.C. Bosman, J.L. Hurink, G.J.M. Smit, Simulation of the effect of introducing micro-generation, energy buffers and accompanied optimization algorithms on the energy efficiency, accepted for ProRISC 2008, October 2008
